How mr-homeless/waldo works
Waldo is a local CS2 gameplay cheat-detection application. Its primary workflow takes uploaded footage, extracts killshot clips, trains or fine-tunes a user-labeled model, analyzes selected clips, and presents persistent scores and frame-level evidence for review and export. The graph combines README-described product capabilities with sampled Flask, dataset, engine, and model code; unsampled route wiring and storage details are represented conservatively.
Main entry points: main.py Web application · train_cheater_pred.py Training entrypoints · datasets.py Labeled clips · autoedit_improved.py Audio clipper
Components 19
Application workflow
- Web application
main.py - Saved results
main.py
Footage processing
- Audio clipper
deepcheat/utils/autoedit/autoedit_improved.py - Video cropping
deepcheat/utils/new_crop.py - Killshot detection
deepcheat/VideoMAEv2/eval_killshot_pred.py - Feature extraction
deepcheat/VideoMAEv2/extract_tad_feature.py
Model training
- Labeled clips
deepcheat/VideoMAEv2/dataset/datasets.py - Dataset assembly
deepcheat/VideoMAEv2/dataset/build.py - Masked video model
deepcheat/VideoMAEv2/models/modeling_pretrain.py - Pretraining engine
deepcheat/VideoMAEv2/engine_for_pretraining.py - Classifier model
deepcheat/VideoMAEv2/models/modeling_finetune.py - Fine-tuning engine
deepcheat/VideoMAEv2/engine_for_finetuning.py - Training entrypoints
deepcheat/VideoMAEv2/train_cheater_pred.py
Clip analysis
- Clip inference data
deepcheat/VideoMAEv2/dataset/datasets.py - Clip scoring
deepcheat/VideoMAEv2/eval_mae.py - Killshot evaluation
deepcheat/VideoMAEv2/eval_killshot_pred.py
Other components
- Local user
- Gameplay footage
- Trained models
Connections 24
- Local user → Web application: uploads and selects
- Web application → Audio clipper: dispatches processing
- Audio clipper → Gameplay footage: reads audio/video
- Audio clipper → Labeled clips: extracts clips
- Web application → Video cropping: dispatches cropping
- Video cropping → Labeled clips: writes processed clips
- Web application → Training entrypoints: starts training
- Training entrypoints → Dataset assembly: builds training data
- Dataset assembly → Labeled clips: loads labeled clips
- Training entrypoints → Pretraining engine: runs pretraining
- Pretraining engine → Masked video model: optimizes
- Training entrypoints → Fine-tuning engine: runs fine-tuning
- Fine-tuning engine → Classifier model: optimizes
- Classifier model → Trained models: saves checkpoints
- Local user → Web application: requests analysis
- Web application → Killshot detection: dispatches killshot analysis
- Killshot detection → Gameplay footage: reads gameplay
- Web application → Clip inference data: selects clips
- Clip inference data → Labeled clips: loads frames and labels
- Web application → Clip scoring: dispatches scoring
- Clip scoring → Clip inference data: reads clip batches
- Clip scoring → Classifier model: loads model
- Web application → Saved results: saves analysis
- Local user → Saved results: reviews and exports